It seems the training data was almost certainly built from adult actress portraits of which you can easily scrape endless amounts from, like any porn site.
I'm not an artist and I still can't wait to get a closer look at their new geometry renderer tech. The UE5 early access branch is already on github, this is gonna be wild. I also saw they are planning to deprecate raytracing and... use lumen for everything? Wonder what Nvidia says to that. But with the next-gen consoles not using RTX cards I guess this is a reasonable move by epic.